Single cell density prediction based on optically induced electrokinetics (OEK) and machine learning

Single cell density is a key indicator for judging cell physiological state, crucial for studying cell function. However, existing measurement methods are often complex and time-consuming, limiting their efficiency in practical applications. To address this, we developed a machine learning-driven single cell density prediction system based on an optically induced electrokinetics (OEK) platform. First, the OEK platform was designed to enable non-invasive electrical manipulation of cells, and cell motion trajectories were obtained using a Depth-from-Defocus (DFD)-based template matching algorithm. Then, the time series of matched frame counts during sedimentation were extracted to characterize feature differences among cells with varying densities. Finally, Bayesian optimization was applied to a gradient boosting machine (GBM) model for parameter tuning and density prediction. The proposed method achieves an R² of 0.950, a root mean square error (RMSE) of 0.0037 g cm^-3, and a mean absolute error (MAE) of 0.0028 g cm^-3, yielding the lowest prediction errors compared with several mainstream machine learning models and reducing computation time and load compared to our previous method. These results demonstrate the effectiveness of the proposed method, which is expected to improve measurement efficiency and offer a new tool for cell biomedical research.